SOM-BP Neural Network-Based Financial Early-Warning for Listed Companies

To prevent and reduce corporate financial risks, this paper builds a financial early-warning model for listed companies based on a combination of SOM and BP neural networks focusing on short-term financial forecasting and monitoring. Firstly, SOM network is utilized to allow self-modification of unit connection weights according to the feature information of input data and enable the weight vector distribution to be similar to the distribution of sample data, thereby obtaining relatively optimal training samples among all training samples. Then, a short-term financial early-warning monitoring model is created through iterative BP training with the relatively optimal samples extracted as the input information of the BP neural network model. The results show that the proposed financial earlywarning system has higher recognition accuracy than the direct use of Logistic model, BP model or SVM model in term of short-term forecasting and monitoring. Furthermore, our model requires less amount of data while ensuring the validity. Therefore, it can monitor financial crises in real time for listed companies, so as to effectively prevent and resolve their financial risks and crises.